Download Revealing the architecture of gene regulation: the promise of eQTL

Review
Revealing the architecture of gene
regulation: the promise of eQTL
studies
Yoav Gilad1, Scott A. Rifkin2 and Jonathan K. Pritchard1
1
2
Department of Human Genetics, University of Chicago, Chicago, IL 60637, USA
Department of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
Expression quantitative trait loci (eQTL) mapping studies have become a widely used tool for identifying
genetic variants that affect gene regulation. In these
studies, expression levels are viewed as quantitative
traits, and gene expression phenotypes are mapped to
particular genomic loci by combining studies of variation
in gene expression patterns with genome-wide genotyping. Results from recent eQTL mapping studies have
revealed substantial heritable variation in gene expression within and between populations. In many cases,
genetic factors that influence gene expression levels can
be mapped to proximal (putatively cis) eQTLs and, less
often, to distal (putatively trans) eQTLs. Beyond providing great insight into the biology of gene regulation, a
combination of eQTL studies with results from
traditional linkage or association studies of human disease may help predict a specific regulatory role for
polymorphic sites previously associated with disease.
The study of gene expression phenotypes
Variation in gene expression is abundant in all organisms
studied to date [1–3]. It has been suggested repeatedly that
modifications in gene regulation are responsible for much
of the observed phenotypic variation in natural populations. Indeed, like substitutions at the protein level,
changes in gene regulation have been found to underlie
numerous adaptive phenotypes in a variety of organisms,
from beak morphology in Darwin finches [4], bristle number, wing pigmentation and trichome patterns in fruit flies
[5–7], branching structure in maize [8], skeletal patterning
and pelvic reduction in sticklebacks [9,10] to parental care
in rodents [11]. Moreover, mutations in putative regulatory
regions have been associated with > 100 human phenotypes including diverse aspects of behavior, physiology and
disease (for a review, see Refs. [12,13]). Despite accumulating evidence that regulatory changes contribute to many
important phenotypes, we still know little about the architecture of gene regulation (see Glossary) or about the
genetic basis for variation in gene expression levels.
In particular, although we understand how mutations
in coding regions affect the amino acid composition of
proteins and, sometimes, how these mutations lead to
differences in phenotypes, the effect of variation at the
Corresponding authors: Gilad, Y. ([email protected]); Rifkin, S.A.
([email protected]); Pritchard, J.K. ([email protected]).
408
DNA level on transcript abundance remains elusive. In fact,
it is difficult to identify regulatory regions in the genome, let
alone to predict how polymorphisms in regulatory regions
affect gene expression levels temporally or spatially [13].
This task becomes particularly important in humans
because many of the loci identified in recent genome-wide
association studies of complex human diseases are located
outside of coding regions and hence are expected to have a
function in gene regulation (e.g. Refs. [14–17]).
Expression quantitative trait loci (eQTL) mapping is
one approach to determine which genomic regions help
to regulate transcription and to study the impact of
polymorphisms within these regions. In such studies, gene
expression levels are treated as quantitative traits, and
their genetic basis can be studied using well-established
linkage and association mapping tools (Figure 1; Box 1).
However, unlike traditional QTL mapping, which is typically limited to a few quantitative traits (e.g. Refs. [18,19]),
DNA microarrays make it possible to measure the expression phenotypes of most genes in a genome simultaneously
and map these phenotypes to specific genomic regions
using genome-wide genetic markers.
Genome-wide mapping of eQTLs can provide great
insight into the genetic architecture of gene expression
Glossary
Expression quantitative trait loci (eQTL) hotspot: a locus in which genetic
variation is associated with the expression variation of many genes. Because
the resolution of the mapping depends on the density of markers, an eQTL
hotspot may reflect the presence of a single influential regulator (such as a
transcription factor) or several linked loci that affect transcript levels of
different genes.
Genetic architecture of a quantitative trait: a description of the association
between variation at the DNA sequence level and variation in a quantitative
trait (e.g. variation in gene expression). Based or the patterns of genetic
association with variation in the quantitative trait, the genetic architecture is
classified as single or multilocus traits, which can interact additivity, or include
epitasis, dominance, cis and/or trans effects.
Heritability: Heritability is the phenotypic variance in the population that is
caused by genetic variation, divided by the total phenotypic variance.
Heritability is usually estimated using the extent of similarity among relatives;
however, most studies cannot exclude the possibility that environmental
factors that are shared among relatives might inflate heritability estimates.
Linkage disequilibrium (LD): LD refers to the situation in which the alleles at
one marker tend to co-segregate with particular alleles at a second marker. For
example, in an extreme case, the ‘A’ allele at one single nucleotide
polymorphism (SNP) might always appear together with the ‘B’ allele at a
second SNP on the same chromosome, and similarly, the alternative alleles ‘a’
and ‘b’ might always co-occur. Strong LD usually occurs only between pairs of
markers that are separated by less than a few tens of kilobases.
0168-9525/$ – see front matter ! 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.tig.2008.06.001 Available online 1 July 2008
Review
Trends in Genetics
Vol.24 No.8
Figure 1. Example of an expression quantitative trait loci (eQTL) for the HLA-C gene in the HapMap European samples (data from Ref. [55]). (a) Plot of !log(P values) for the
association between individual single nucleotide polymorphisms (SNPs) and expression of HLA-C. The location of the gene is indicated by the small red bar at the bottom of
the figure, and the x-axis measures location relative to the transcription start site (TSS). Each data point is for a single SNP. (b) Individual expression levels of HLA-C,
grouped according to the genotype of the most significant SNP in the region (rs2249741; indicated by the red data point in panel (a). Interestingly, one of the SNPs in the
signal peak in panel (a) (rs92644942) has also been associated with HIV set point, suggesting that higher expression of HLA-C can help reduce HIV viral load [62].
variation because it simultaneously captures many regulatory interactions. The long-term goal of eQTL mapping
studies is to elucidate how genotypic variation underlies
morphological or physiological consequences by using gene
expression levels as intermediate molecular phenotypes.
For example, by combining eQTL mapping with results
from traditional linkage or association studies of human
disease, one can assign a specific regulatory role to polymorphic sites in a genomic region known to be associated
with disease (e.g. Ref. [20]).
Here we discuss the general principles of regulatory
mechanisms that are emerging from recent QTL mapping
efforts in humans and other organisms and explore the
challenges of mapping regulatory variation in different
species. Perhaps the most salient findings of eQTL studies
thus far are that (i) variation in gene expression levels is
both widespread and highly heritable; (ii) gene expression
levels are highly amenable to genetic mapping and (iii)
most strong eQTLs are found near the target gene,
suggesting that variation in cis regulatory elements
underlies much of the observed variation in gene expression levels.
Variation in gene expression is widespread in human
populations
Gene expression levels measured by microarrays can
be affected by many nongenetic factors, including environmental variation, epigenetic modifications and random
fluctuations in expression, as well as by experimental
issues including measurement error, staging and (for many
of the studies) variation that arises in transformed cell
lines [21]. For that reason, it was not clear initially how
heritable these measured expression levels would be, and
early eQTL studies spent considerable effort addressing
this basic question [22–25]. Indeed, measured gene expression levels of most genes were found to have statistically
significant heritability.
For example, Göring et al. [24] analyzed expression data
for lymphocytes isolated from 1240 individuals representing 30 large families. After removing genes whose expression levels were below a baseline threshold established by
negative control samples, they found that 86% of probes
mapping to RefSeq genes showed significantly heritable
expression levels [at a false discovery rate (FDR) of 1%].
However, it is important to note that the actual level of
409
Review
Box 1. Expression quantitative trait loci mapping methods
Mapping approaches for expression quantitative trait loci (eQTLs) in
humans and other natural populations can be classified into linkage
methods and association methods (reviewed in Ref. [63]). Briefly,
linkage mapping uses a study design that is based on tracking the
transmission of chromosomes through families. This approach aims
to identify markers, or chromosomal segments, whose transmission
patterns are correlated with the phenotype – implying that they are
linked to QTLs. By contrast, association mapping, in its simplest
form, uses samples of unrelated individuals. Here the goal is to
identify markers whose genotype is correlated with the phenotype
of interest, again implying that those markers are linked to QTLs.
The major advantage of linkage mapping is that a genome-wide
scan can be performed using small numbers of markers [e.g. <1000
microsatellites, or slightly larger numbers of single nucleotide
polymorphisms (SNPs), are usually sufficient for linkage mapping
in humans]. However, for detecting common variants that affect
gene expression, association mapping is a far more powerful
approach [64], provided that the causal variants are in strong
linkage disequilibrium (LD) with genotyped SNPs. Hence, with
sufficiently dense genotyping, association mapping is much more
likely to identify eQTLs with small or medium effect sizes. One
plausible concern for association mapping is the possibility of false
positives owing to population structure [65,66]. However, this issue
can be overcome for most eQTL studies by applying recently
developed methods for using genome-wide SNP data to correct for
population structure [67].
Association mapping can also provide fine-scale resolution on the
locations of functional variants (usually within a few tens of kb in
humans, depending on the local extent of linkage disequilibrium).
By contrast, linkage mapping provides much more coarse-grained
localization because one relies on the occurrence of recombination
events within the pedigrees to help fine map the relevant variants.
Now that high-density genome-wide genotyping is readily available,
association mapping will likely be the method of choice for future
eQTL studies.
Study designs in model organisms, such as yeast, flies and mice,
often share characteristics of both linkage and association mapping.
For example, Brem and Kruglyak [45] created 40 haploid yeast
strains that were segregants from a cross between two parental
strains. They wanted to see if there was any correlation between
expression levels and marker genotypes (the latter effectively
identify the parental origin of the chromosomal segment). Their
approach can be viewed as either linkage (correlation between
transmitted chromosomes and phenotype) or association (correlation between genotype and phenotype). Because this type of design
allows one to test directly for association between genotype and
phenotype, it achieves the statistical power of an association study;
however, it usually provides poor localization of the associated
variants because the numbers of recombination events are very
limited [45].
Finally, allele-specific expression assays (e.g. Refs. [27,68]) offer a
fundamentally different approach to discovering factors that might
affect gene expression levels. In these studies, one tests whether the
two copies of a gene from a single individual are expressed in equal
amounts. Assuming that both chromosomes are exposed to the same
soup of trans-acting factors, a difference between the expression
levels of the two copies implies that there is a functional difference in
cis between the two chromosomes. This type of approach can, in
principle, identify effects that would not be detected by standard eQTL
methods: for example, variation that is present in just a single
member of a sample or noninherited epigenetic factors. However, the
method provides no guidance as to the location of the causal
variation, beyond the implication that it acts in cis.
heritability is moderate for most of these genes. For
example, 41% of the RefSeq probes had estimated heritability >0.3, but just 5% had heritability >0.5. These
results suggest that, although measured mRNA levels of
most expressed genes are indeed correlated across family
members, nongenetic factors are also likely to be important
410
Trends in Genetics Vol.24 No.8
in determining expression levels. Moreover, because heritability estimates cannot distinguish between familial
correlations due to shared genetic factors as opposed to
correlations caused by shared environment, the environmental contribution could be somewhat larger than the
heritability estimates would imply.
Recently, several groups have also found that, for a large
fraction of loci, mean expression levels vary among populations [24,26,27]. For instance, Stranger et al. [28] estimated that 17–29% of loci have significant differences in
mean expression levels between pairs of HapMap populations. One possible explanation for this observation is
that these expression differences are caused by polymorphic sites with divergent allele frequencies between
the HapMap populations. However, given that few single
nucleotide polymorphisms (SNPs) in the genome have
large frequency differences between populations [29], it
seems likely that much of the expression variation across
populations is caused by environmental factors. Indeed, it
has been shown that even very closely related populations
living in different environments can have substantially
different expression profiles. For example, Idaghdour
et al. [30] studied gene expression in leukocyte samples
from Moroccans living in three different environmental
conditions: in a city, in a mountain village and in a desert.
Although these three groups are similar at the genetic
level, the authors estimated that 37% of expressed genes
show significant differences in mean expression levels
among the three groups (at an FDR of 1%). Additionally,
technical differences in the preparation or propagation of
samples might also create apparent differences in gene
expression between populations (as has been suggested for
the much older CEU cell lines compared with the other
HapMap samples [28]).
Cis and trans regulation of gene expression
A common observation, from a variety of both linkage and
association eQTL studies, is that numerous genes
have proximal eQTLs, likely in cis regulatory elements
(see Box 2 for a discussion of the definitions of cis and
trans). For example, Stranger et al. [28] measured gene
expression in the transformed lymphoblastoid cell lines
that were prepared by the International HapMap Project
[29]. Using 2.2 million common SNPs genotyped by the
HapMap to test for association in 210 of these cell lines
(derived from unrelated individuals), they identified 831
genes with a significant proximal eQTL (at an FDR of 5%;
proximal regions were defined as a 2-MB window containing the gene). More recently, Emilsson et al. [25] used
microsatellites to map eQTLs and identified proximal
eQTLs for 9% of genes in blood and 6% of genes in adipose
tissue after applying linkage analysis to samples of 938
and 570 individuals, respectively (proximal signals in this
case were defined as signals that were significant at the
nearest microsatellite to the gene).
The precision of localization of eQTLs in associationbased studies is limited by the extent of strong linkage
disequilibrium (LD) and often provides resolution to within
10–20 kb in humans. Two studies, using different samples
of transformed lymphoblastoid cell lines, concluded that
most proximal eQTLs lie close to the actual genes [23,28].
Review
Box 2. Transcriptional regulation: cis and trans elements
The terms cis and trans were introduced to genetics by Haldane [69]
to describe differences in the configurations of mutant alleles in
heterozygotes, in analogy with cis and trans isomers in chemistry.
The terms cis and trans were essentially replacements for Bateson’s
terms ‘coupling’ and ‘repulsion’. In the cis (coupling) configuration,
two mutations were inherited together, whereas in the trans
(repulsion) configuration, the two mutations were found on different
members of a pair of homologous chromosomes. Morgan et al. [70]
proposed that linkage between the mutations was responsible for
unequal numbers of the two types of heterozygotes in crosses, and
Lewis [71] operationalized the definition in his cis-trans test for
position effects, which has been widely used as a method for
detecting whether two mutants lie in the same gene.
Molecular studies of gene regulation have classified regulatory
interactions based on their effects in cis or trans, and the terms have
been co-opted to describe particular types of regulatory elements.
Consistent with the original definitions, cis regulatory elements
have an allele-specific effect on gene expression, whereas trans
elements affect the regulation of both alleles. Examples of trans
elements may be transcription factors or insulators that regulate
transcription initiation or small interfering RNA that regulates RNA
stability. Examples of cis elements include promoter regions,
enhancers and boundary elements, which regulate transcription
initiation, or poly-A signals and siRNA binding sites, which regulate
RNA stability [72].
In the expression quantitative trait loci (eQTL) mapping literature,
regulatory polymorphisms are often said to be in cis or in trans on
the basis of their physical distance from the regulated gene.
Regulatory variation that is mapped near the target gene is
classified as being cis. Although such a distance-based definition
probably provides a reasonable broad classification of cis and trans
regulatory elements (e.g. the ENCODE project found that "60% of
transcription factor–binding sites – one type of cis regulatory
elements, reside within 3 kb of transcription start sites [57]), this
definition becomes problematic if trans regulatory elements exist
near their gene target (e.g. a transcription factor that regulates
adjacent target genes). Alternatively, the distance-based definition
of regulatory elements might lead to the misclassification of long
distance cis elements as ‘trans’ owing to conservative definitions of
the proximal ‘cis windows’ (e.g. defining as cis only eQTLs that
reside within 100 kb of the transcription start site). In any case,
distance-based classification of eQTLs as cis or trans requires
empirical validation.
For example, data from Stranger et al. [28] indicate that
most of the proximal eQTLs map within "100 kb of the
transcription start site of the regulated gene. Similarly,
Dixon et al. [23] found only a few eQTLs that were located
>100 kb from the relevant gene on the same chromosome.
In addition, they observed that more distal eQTLs were
usually much weaker than were signals close to the gene.
Given that LD tends to spread signals out, these observations argue that most determinants of gene regulation
tend to be close to the target gene, putatively in cis, and
that long-range regulators are either much less frequent or
exert much smaller effects. Consistent with this conjecture, distal-acting eQTLs were found to have much smaller
effect sizes compared with proximal eQTLs (e.g. Ref. [23]).
The enrichment of proximal eQTLs that are near
the genes they regulate is consistent with the view
that changes in cis regulatory elements are less likely
to have deleterious effects than changes in trans,
because mutations in cis elements are more likely to
affect the regulation of only one gene [13,26]. Thus, these
observations suggest that genetic variation in cis regulatory elements might have a disproportional effect on gene
Trends in Genetics
Vol.24 No.8
expression variation and perhaps on gene expression evolution [27].
However, the enrichment of proximal compared with
distal eQTLs might be overestimated because of statistical
and technical reasons. For example, it should be noted that
it is more difficult to detect a distal eQTL than a proximal
eQTL of the same effect-size because the tests for distal
effects are subject to a much greater burden of multiple
testing (because, whereas the window size for proximal
eQTLs is small, distal eQTLs can be found anywhere in the
genome). Consistent with the view that many distal eQTLs
are missed, at the time of writing, the human data do not
seem to indicate strong clustering of distal-acting eQTLs
into ‘master regulators’ as reported in other organisms
inclduing mice and Drosophila [23,25] (although possible
examples in humans have been reported [28,31]).
In addition, sequence polymorphisms at the microarray
probes that are used to measure gene expression might be
responsible for some of the observed proximal eQTLs.
Because sequence mismatches between the target and
the probe affect microarray hybridization intensity [32],
one might estimate different expression levels for samples
with different alleles at the probe sequence. These apparent differences in gene expression will be associated with
any marker that is in LD with polymorphisms in the probe
and thus will be mapped as spurious proximal eQTLs. The
number of proximal eQTLs that can be explained by this
technical artifact is unknown. Alberts et al. [33] estimated
that 24% of probe sets on Affymetrix human gene expression arrays contain a SNP in at least one probe (using the
HapMap SNPs) and that 4% of probe sets contain a SNP in
three or more probes. These numbers are likely to be
underestimates of the true proportion of probes that contain SNPs, because current databases of human variation
such as HapMap and dbSNP still do not contain many of
the common variants in the genome [34]. For that reason,
excluding probes with known SNPs can is only a partial
solution to this problem. To date, however, most eQTL
studies do not even exclude probes with known SNPs.
Mapping in model organisms: evidence for eQTL
hotspots
Genome-scale eQTL mapping studies in nonhuman organisms have predominantly focused on three objectives: (i) to
identify QTLs associated with variation in transcript abundances in defined mapping populations and categorize
them as proximal or distal to the locus of the transcript
they affect, (ii) to determine the numbers, genomic distributions and magnitudes of eQTL effects on transcript
levels and (iii) to evaluate whether eQTLs interact additively to control transcript levels. Despite differences in
experimental designs and analysis methods, the relatively
young eQTL mapping literature suggests some basic
answers, at least with respect to the first two objectives.
Unlike in humans, populations with defined genetic
relationships can be constructed in model organisms. Such
study designs increase the statistical power of QTL studies
[35] but can decrease their generality because a restricted
subset of alleles ultimately segregate in the mapping
population, leaving many potentially consequential loci
monomorphic. There are three commonly used designs
411
Review
Trends in Genetics Vol.24 No.8
Figure 2. Breeding designs commonly used in model organism expression quantitative trait loci (eQTL) studies. The original inbred lines are shaded in different colors to
track transmission of segments from one pair of homologous chromosomes. However, as inbred lines 1 and 2 undoubtedly are not polymorphic for many alleles, the actual
genotypes at many marker loci on the grey and black chromosomes will be the same. (a) F2 design. Two inbred lines are crossed to form a heterozygous but identical F1
generation. These F1 individuals are then crossed to form an F2 generation. (b) Backcross design. Two inbred lines are crossed to form an F1 generation. F1 individuals are
then backcrossed to one of the inbred parents to form a B1 generation. (c) Recombinant inbred lines. Inbreeding from the F2 generation on eventually results in near
homozygous individuals with a mixture of markers from the original inbred lines. One such line is depicted starting from a brother-sister pair of F2 individuals. Different
recombinant lines can be created by crossing different brother-sister pairs.
in the eQTL literature (Figure 2). In the ‘F2 design’, two
inbred lines are mated to form a heterozygous F1 generation. These F1 organisms are then interbred to form an F2
generation. Independent assortment and recombination
will scramble the genomes in these F2 organisms generating genotypic variation (e.g. Refs. [36–39]). In the ‘backcross design’, the F1 organisms are mated to one of the two
parents to form a B1 generation [40]. For nonclonal organisms, F2 and B1 mapping populations are ephemeral. To
create a permanent mapping population, F2 individuals
can be selfed or brother-sister mated for several generations to create a set of ‘recombinant inbred lines’. Such
lines are largely homozygous but have distinct combinations of parental alleles [3,39,41,42].
The power to identify loci associated with transcriptional variation depends in part on the size of the mapping
population and the size of the effect of allelic variation on
transcript abundances. As with most statistical procedures, larger sample sizes yield better estimates, but,
because of the expense of genome-wide measurements,
most eQTL mapping studies in model organisms analyze
relatively few lines!almost always <100. This severely
limits the ability to detect eQTLs with small effects,
particularly when significance levels are adjusted for
multiple testing. Because the number of tests performed
for distal eQTLs is much greater than for proximal eQTLs
(as explained earlier), and because eQTL mapping studies
routinely find that the magnitudes of the effects of proximal (putatively cis) eQTLs are larger than those that
412
affect distant (putatively trans) loci [21,43,44], the true
ratio of putative trans to putative cis eQTLs is difficult to
estimate.
Despite low power to identify eQTLs with small effects,
transcript abundances are generally found to be polygenic
traits [43,45,46]. Moreover, in contrast to current results in
humans, expression QTLs in several studies in model
organisms were found to be unevenly distributed across
chromosomes, with several examples of regions of high
distal eQTL concentration (i.e. eQTL hotspots). Although
some of these might be artifacts of microarray normalization [47], regions with a high number of distal eQTLs
might harbor ‘master regulators’ that affect the expression
levels of many genes. For example, Mehrabian et al. [48]
dissected an eQTL hotspot in mice and identified a locus
that affects the regulation of several metabolic traits
associated with obesity and bone density. Similarly, West
et al. [43] identified several genomic hotspots in Arabidopsis in which different alleles were associated with the
regulation of a large number of genes. Interestingly,
transcription factors were not overrepresented in the
hotspots, reinforcing the idea that changes in any part
of a network could percolate through to affect many other
genes.
Beyond eQTL screens in model organisms?
One of the long-term goals of eQTL studies in model
organisms is to delineate which eQTLs regulate metabolic
and developmental pathways [49] – not an easy task, given
Review
that gene expression is often a polygenic trait with genes of
both major and minor effects. Furthermore, transcriptional networks can be plastic [50,51] and, as a consequence, eQTLs can be extremely context dependent,
differing because of a large number of factors, such as
temperature [41], sex [38], developmental stage and tissues [44]. The task of mapping entire regulatory networks
given these difficulties seems daunting [52].
One other complication of eQTL mapping studies in
model organisms is that they predominantly use populations derived from two inbred lines. Quantitative genetic
estimates are only applicable to the population from which
they are derived and depend fundamentally on causative
genetic variation being present in the population. The
ranges of expression levels in mapping populations often
fall outside the parental range, suggesting rampant
genetic interactions including potential epistatic and/or
compensatory effects. The use of constructed populations
is typically a great advantage of working with model
organisms, but in this case, it limits the applicability of
eQTL mapping results. Distal eQTLs are difficult to replicate even from independent crosses derived from the same
parents [39], and to date, no studies have attempted to
replicate eQTLs using crosses of different genotypes. Transcriptional networks in natural populations, including
humans, operate in much richer genotypic and environmental contexts.
In contrast to studies in model organisms, eQTLs are
being mapped in humans using natural populations. Thus,
although eQTL studies in humans face the same challenges regarding the plasticity of the polygenic expression
phenotypes, results obtained in human studies are not
expected to be restricted only the samples used (when
population structure is taken into account).
Concluding remarks and future perspectives
The considerable advances in expression quantitative trait
loci (eQTL) studies notwithstanding, there are still open
questions about the biology and applications of eQTL
mapping. First, there are important technical questions
about the extent to which eQTLs are replicated across
independent samples and independent platforms for
measuring gene expression (see Box 3 for a discussion
on different approaches to compare results across studies).
Second, most of the human eQTL studies to date have
analyzed transformed lymphoblast cell lines or lymphocyte
samples, because these are the most readily available
tissues [22–24,28,31,53–55]. However, because expression
patterns differ dramatically across tissues, there is now
great interest in collecting similar data for a much wider
variety of tissues. Indeed, three recent studies have started
moving beyond cell types in blood by characterizing eQTLs
in cortical [56], adipose [25] and liver [20] tissues.
Expression QTL mapping can provide great insight into
the biology of gene regulation. One can view eQTL mapping as a sort of large-scale mutagenesis experiment, in
which "10 million common single nucleotide polymorphisms (SNPs) have been sprinkled down on the human
genome, and each individual receives a random collection
of these. Measurements of gene expression provide us with
a tool for learning which types of SNPs are most likely to
Trends in Genetics
Vol.24 No.8
Box 3. How to compare results across expression
quantitative trait loci studies?
Although several robust patterns emerge when results from multiple expression quantitative trait loci (eQTL) studies are considered,
one discouraging observation is that specific eQTLs are not
generally replicated across studies [24,39]. For example, Göring
et al., [24] reported that, although they replicated 11 of the top 13
proximal linkage signals from Morley et al., [31], they failed to
replicate even the top distal signals from Morley et al. The
discrepancy can partly be explained by the overall low power to
detect eQTLs, particularly distal ones. However, an alternative (yet
related) explanation for the lack of replication might be the method
used to compare results across studies. In most studies, eQTLs are
identified using an arbitrary statistical cut-off to ensure a minimum
number of false positives. This approach leads to numerous false
negatives, which might increase the discrepancies between the lists.
For example, consider a true eQTL with small effect that has been
detected as significant in one study (e.g. with P < 0.05) but did not
reach the arbitrary statistical cut-off in another study (e.g. P = 0.06).
An effective comparison of such two studies would not consider
these results to be a discrepancy.
A second problem is that significant eQTLs are classified as cis or
trans based on an arbitrary distance cut-off relative to the regulated
gene (see Box 2). Typically, the results are analyzed by comparing the
lists of significant cis and trans eQTLs [24,73]. However, as pointed
out by Williams et al., [73], a comparison of such lists is confounded
by the different arbitrary decisions made in each study. Thus, a metaanalysis of the data from multiple studies, using a single, consistent
method, is necessary for a meaningful comparison.
One approach for an effective comparison between studies is to
estimate the power to detect any eQTL that is seen in one study
using the parameters of an alternative study. Instead of comparing
the entire lists of significant eQTLs in both studies, one would focus
only on the eQTLs for which the lack of replication is inconsistent
with the estimated statistical power [39].
Alternatively, one could use a statistical model to simultaneously
analyze the data from multiple studies. Such a statistical model
would be designed to estimate the proportion of true eQTLs across
studies, taking into account that eQTLs that are identified as
significant in at least one dataset are more likely to be eQTLs in
other datasets. Such an approach is similar to methods for
controlling the false discovery rate in microarray experiments, by
estimating from the data the proportion of genes that are
differentially expressed.
affect gene regulation, in a way that complements other
experimental approaches such as allele-specific expression
studies [57] or reporter gene assays [58].
We believe that this type of data can yield much more
fine-grained information about the impact of individual
SNPs on expression levels. For example, the extent to
which eQTLs will be shared across diverse tissues is still
unknown. Similarly, eQTL data have yet to be used to infer
regulatory networks. This could be done, for example, by
identifying proximal eQTLs for transcription factors that
also mapped as distal eQTLs for other genes (implying a
regulatory interaction between the transcription factor
and its target). To achieve the potential of eQTL mapping
as a tool for understanding gene regulation, it will be
necessary to fine map the functional eQTL sites. For some
purposes, we might simply want to estimate an approximate physical location of the causal variant(s). However, it
will often be of interest to use the data to shortlist those
variants that might be the actual functional sites [52] and
to proceed to functional validation. It is usually difficult to
identify functional variants in humans with any confidence
in silico, because the HapMap – the primary genome-wide
413
Review
resource on genotype data in humans – currently contains
only approximately one third of common SNPs [34] and is
therefore expected to contain only a minority of the
relevant functional variants. By contrast, the forthcoming
1000 Genomes Project (http://www.1000genomes.org/) will
soon provide resequencing data on large numbers of individuals, thus providing reasonably complete information
on common variation throughout the genome (outside
highly repetitive regions). These new data, along with
new imputation methods (e.g. Refs. [52,59]), should accelerate the process of identifying the functional alleles that
affect gene expression levels.
Finally, eQTL mapping can provide important information for dissecting the genetics of complex disease
(Figure 1). In its simplest form, the identification of eQTLs
can provide a tool for connecting SNPs that are significant
in genome-wide association studies of disease to a molecular mechanism [14,20]. Expression QTLs may also be
useful for linking genes and individual variants to cellular
phenotypes, such as cell line sensitivity to chemotherapeutic agents [60]. More ambitiously, one might be able to use
patterns of gene expression and eQTL mapping in people
with and without disease to identify networks of genes that
are differentially regulated in the two groups [61]. Any
eQTL that up- or downregulates such a network is a
natural candidate for affecting the disease phenotype itself
and would be of particular interest in association studies of
the disease.
Acknowledgements
The authors thank J. Borevitz for helpful discussions and Sridhar
Kudaravalli for helping to generate Figure 1. Y.G. is supported by NIH
Grant GM077959, J.K.P. is supported by Grant HG002772 and S.A.R. is
supported by NIH fellowship 5F32GM080966.
References
1 Oleksiak, M.F. et al. (2002) Variation in gene expression within and
among natural populations. Nat. Genet. 32, 261–266
2 Gilad, Y. et al. (2006) Natural selection on gene expression. Trends
Genet. 22, 456–461
3 Genissel, A. et al. (2008) Cis and Trans Regulatory effects contribute to
natural variation in transcriptome of Drosophila. Mol. Biol. Evol. 25,
101–110
4 Abzhanov, A. et al. (2004) Bmp4 and morphological variation of beaks
in Darwin’s finches. Science 305, 1462–1465
5 McGregor, A.P. et al. (2007) Morphological evolution through multiple
cis-regulatory mutations at a single gene. Nature 448, 587–590
6 Stern, D.L. (1998) A role of Ultrabithorax in morphological differences
between Drosophila species. Nature 396, 463–466
7 Gompel, N. et al. (2005) Chance caught on the wing: cis-regulatory
evolution and the origin of pigment patterns in Drosophila. Nature 433,
481–487
8 Clark, R.M. et al. (2006) A distant upstream enhancer at the maize
domestication gene tb1 has pleiotropic effects on plant and inflorescent
architecture. Nat. Genet. 38, 594–597
9 Shapiro, M.D. et al. (2004) Genetic and developmental basis of
evolutionary pelvic reduction in threespine sticklebacks. Nature 428,
717–723
10 Cresko, W.A. et al. (2004) Parallel genetic basis for repeated evolution
of armor loss in Alaskan threespine stickleback populations. Proc.
Natl. Acad. Sci. U. S. A. 101, 6050–6055
11 Hammock, E.A. and Young, L.J. (2005) Microsatellite instability
generates diversity in brain and sociobehavioral traits. Science 308,
1630–1634
12 Kleinjan, D.A. and van Heyningen, V. (2005) Long-range control of
gene expression: emerging mechanisms and disruption in disease. Am.
J. Hum. Genet. 76, 8–32
414
Trends in Genetics Vol.24 No.8
13 Wray, G.A. (2007) The evolutionary significance of cis-regulatory
mutations. Nat. Rev. Genet. 8, 206–216
14 Moffatt, M.F. et al. (2007) Genetic variants regulating ORMDL3
expression contribute to the risk of childhood asthma. Nature 448,
470–473
15 Wellcome Trust Case Control Consortium (2007) Genome-wide
association study of 14,000 cases of seven common diseases and
3,000 shared controls. Nature 447, 661–678
16 Easton, D.F. et al. (2007) Genome-wide association study identifies
novel breast cancer susceptibility loci. Nature 447, 1087–1093
17 Helgadottir, A. et al. (2007) A common variant on chromosome 9p21
affects the risk of myocardial infarction. Science 316, 1491–1493
18 Robin, C. et al. (2002) hairy: A quantitative trait locus for drosophila
sensory bristle number. Genetics 162, 155–164
19 Macdonald, S.J. et al. (2005) The effect of polymorphisms in the
enhancer of split gene complex on bristle number variation in a
large wild-caught cohort of Drosophila melanogaster. Genetics 171,
1741–1756
20 Schadt, E.E. et al. (2008) Mapping the genetic architecture of gene
expression in human liver. PLoS Biol. 6, e107
21 Gibson, G. and Weir, B. (2005) The quantitative genetics of
transcription. Trends Genet. 21, 616–623
22 Cheung, V.G. et al. (2003) Natural variation in human gene expression
assessed in lymphoblastoid cells. Nat. Genet. 33, 422–425
23 Dixon, A.L. et al. (2007) A genome-wide association study of global gene
expression. Nat. Genet. 39, 1202–1207
24 Göring, H.H. et al. (2007) Discovery of expression QTLs using largescale transcriptional profiling in human lymphocytes. Nat. Genet. 39,
1208–1216
25 Emilsson, V. et al. (2008) Genetics of gene expression and its effect on
disease. Nature 452, 423–428
26 Carroll, S.B. et al. (2004) From DNA to Diversity: Molecular Genetics
and the Evolution of Animal Design. Blackwell Science
27 Wittkopp, P.J. et al. (2008) Regulatory changes underlying expression
differences within and between Drosophila species. Nat. Genet. 40,
346–350
28 Stranger, B.E. et al. (2007) Population genomics of human gene
expression. Nat. Genet. 39, 1217–1224
29 International HapMap Consortium (2005) A haplotype map of the
human genome. Nature 437, 1299–1320
30 Idaghdour, Y. et al. (2008) A genome-wide gene expression signature of
environmental geography in leukocytes of Moroccan Amazighs. PLoS
Genet. 4, e1000052
31 Morley, M. et al. (2004) Genetic analysis of genome-wide variation in
human gene expression. Nature 430, 743–747
32 Gilad, Y. and Borevitz, J. (2006) Using DNA microarrays to study
natural variation. Curr. Opin. Genet. Dev. 16, 553–558
33 Alberts, R. et al. (2007) Sequence polymorphisms cause many false cis
eQTLs. PLoS One 2, e622
34 Frazer, K.A. et al. (2007) A second generation human haplotype map of
over 3.1 million SNPs. Nature 449, 851–861
35 Lynch, M. and Walsh, B. (1998) Genetics and Analysis of Quantitative
Traits. Sinauer
36 Damerval, C. et al. (1994) Quantitative trait loci underlying gene product
variation: a novel perspective for analyzing. Genetics 137, 289–301
37 Yvert, G. et al. (2003) Trans-acting regulatory variation in
Saccharomyces cerevisiae and the role of transcription factors. Nat.
Genet. 35, 57–64
38 Bhasin, J.M. et al. (2008) Sex specific gene regulation and expression
QTLs in mouse macrophages from a strain intercross. PLoS One 3,
e1435
39 Peirce, J.L. et al. (2006) How replicable are mRNA expression QTL?
Mamm. Genome 17, 643–656
40 Klose, J. et al. (2002) Genetic analysis of the mouse brain proteome.
Nat. Genet. 30, 385–393
41 Li, Y. et al. (2006) Mapping Determinants of Gene Expression
Plasticity by Genetical Genomics in C. elegans. PLoS Genet. 2, e222
42 Bao, L. et al. (2006) Combining gene expression QTL mapping and
phenotypic spectrum analysis to uncover gene regulatory
relationships. Mamm. Genome 17, 575–583
43 West, M.A. et al. (2007) Global eQTL mapping reveals the complex
genetic architecture of transcript-level variation in Arabidopsis.
Genetics 175, 1441–1450
Review
44 Petretto, E. et al. (2006) Heritability and tissue specificity of expression
quantitative trait loci. PLoS Genet. 2, e172
45 Brem, R.B. et al. (2002) Genetic dissection of transcriptional regulation
in budding yeast. Science 296, 752–755
46 Brem, R.B. et al. (2005) Genetic interactions between polymorphisms
that affect gene expression in yeast. Nature 436, 701–703
47 Williams, R.W. (2006) Expression genetics and the phenotype
revolution. Mamm. Genome 17, 496–502
48 Mehrabian, M. et al. (2005) Integrating genotypic and expression data in
a segregating mouse population to identify 5-lipoxygenase as a
susceptibility gene for obesity and bone traits. Nat. Genet. 37, 1224–1233
49 Jansen, R.C. and Nap, J.P. (2001) Genetical genomics: the added value
from segregation. Trends Genet. 17, 388–391
50 van Swinderen, B. and Greenspan, R.J. (2005) Flexibility in a gene
network affecting a simple behavior in Drosophila melanogaster.
Genetics 169, 2151–2163
51 Stern, S. et al. (2007) Genome-wide transcriptional plasticity underlies
cellular adaptation to novel challenge. Mol. Syst. Biol. 3, 106
52 Servin, B. and Stephens, M. (2007) Imputation-based analysis of
association studies: candidate regions and quantitative traits. PLoS
Genet 3, e114
53 Cheung, V.G. et al. (2005) Mapping determinants of human gene
expression by regional and genome-wide association. Nature 437,
1365–1369
54 Kwan, T. et al. (2008) Genome-wide analysis of transcript isoform
variation in humans. Nat. Genet. 40, 225–231
55 Stranger, B.E. et al. (2007) Relative impact of nucleotide and copy
number variation on gene expression phenotypes. Science 315, 848–853
56 Myers, A.J. et al. (2007) A survey of genetic human cortical gene
expression. Nat. Genet. 39, 1494–1499
57 ENCODE Project Consortium (2007) Identification and analysis of
functional elements in 1% of the human genome by the ENCODE
pilot project. Nature 447, 799–816
58 Chabot, A. et al. (2007) Using reporter gene assays to identify cis
regulatory differences between humans and chimpanzees. Genetics
176, 2069–2076
Trends in Genetics
Vol.24 No.8
59 Marchini, J. et al. (2007) A new multipoint method for genome-wide
association studies by imputation of genotypes. Nat. Genet. 39, 906–
913
60 Huang, R.S. et al. (2007) A genome-wide approach to identify genetic
variants that contribute to etoposide-induced cytotoxicity. Proc. Natl.
Acad. Sci. U. S. A. 104, 9758–9763
61 Chen, Y. et al. (2008) Variations in DNA elucidate molecular networks
that cause disease. Nature 452, 429–435
62 Fellay, J. et al. (2007) A whole-genome association study of major
determinants for host control of HIV-1. Science 317, 944–947
63 Hirschhorn, J.N. and Daly, M.J. (2005) Genome-wide association
studies for common diseases and complex traits. Nat. Rev. Genet. 6,
95–108
64 Risch, N. and Merikangas, K. (1996) The future of genetic studies of
complex human diseases. Science 273, 1516–1517
65 Devlin, B. and Roeder, K. (1999) Genomic control for association
studies. Biometrics 55, 997–1004
66 Pritchard, J.K. and Rosenberg, N.A. (1999) Use of unlinked genetic
markers to detect population stratification in association studies. Am.
J. Hum. Genet. 65, 220–228
67 Price, A.L. et al. (2006) Principal components analysis corrects for
stratification in genome-wide association studies. Nat. Genet. 38, 904–
909
68 Serre, D. et al. (2008) Differential allelic expression in the human
genome: a robust approach to identify genetic and epigenetic cisacting mechanisms regulating gene expression. PLoS Genet. 4,
e1000006
69 Haldane, J.B.S. (1942) New Paths in Genetics. Harper & Brothers
70 Morgan, T.H. et al. (1915) The Mechanism of Mendelian Heredity.
Henry Holt & Company
71 Lewis, E.B. (1945) The relation of repeats to position effect in
Drosophila melanogaster. Genetics 30, 137–166
72 Wray, G.A. et al. (2003) The evolution of transcriptional regulation in
eukaryotes. Mol. Biol. Evol. 20, 1377–1419
73 Williams, R.B.H. et al. (2007) The influence of genetic variation on gene
expression. Genome Res. 17, 1707–1716
415